Last week, one agent ran autonomously for 17 hours building a website — and another member launched a platform that replaces $50k of subscriptions, live on the call. If you missed it, your agent stack is officially out of date. 😄 📞 HOW THE CALLS WORK The calls can run 2+ hours. We want to make sure we're respecting everyone's time. Especially those of you who actually show up. Here's the structure: 👉 Reply to this post with your questions before the call 👉 If you submit a question and you're on the call, you go first 👉 We work through questions in the order they came in 👉 Then we open it up for everyone else If you can't make the call but want your question answered, drop it in the comments. We'll get to it. But priority goes to people who are there. The goal is simple: if you're taking the time to show up, you shouldn't have to wait behind questions from people who aren't even on the call. There's plenty still in motion: Rod is running comparative benchmarks on GLM 5.3 vs. Claude, Juan's gearing up for his image-to-image venue outreach, and Lisa promised a report back on her harness evaluation (T3, Hermes, OpenRouter). If you've got follow-ups of your own — or questions sparked by last week's personal infrastructure round-robin — drop them below and we'll dig in. 🔗 ZOOM LINK (save this) https://us06web.zoom.us/j/81995207847?pwd=Xe6u6LmIQOmCP5VTnOwWYjDBfZNKGB.1 📅 WHEN Tuesday September 22nd at 6PM ET Looking forward to seeing you on the call!

📝 SUMMARY This week's builder/founder coaching call, led by Brandon Hancock, packed in cost-optimization strategy, agentic architecture patterns, and hands-on go-to-market playbooks. Brandon shared updates on EMS SOAP (slow enterprise sales, active fundraising, aggressive LLM cost-cutting) and his new 30-day challenge project Listio, while Patrick Chouinard gave a deep tour of his personal "agentic OS" built on Proxmox, Hermes, and a markdown-based knowledge graph. The rest of the call was hands-on peer coaching: architecture advice for Hemal's e-commerce AI co-pilot, a full GTM and fundraising playbook for Juan's AI photo booth, cold-outreach troubleshooting for Shakur, and career-positioning strategy for Varun. 💡 KEY INSIGHTS • Treat falling model costs as a strategic weapon: when a model gets 10x cheaper, reinvest the savings into 10x more product value (integrity checks, live QA) rather than pocketing margin. Expect this reset cycle every 6–8 months. • Model swaps can deliver 100x savings: Brandon ran the same classification task on DeepSeek v4 Flash for $3 vs. $350 on a frontier model — same intelligence, fraction of the cost. • A true agent reasons and acts in a loop; a pipeline of sequential LLM calls that just streams an answer is not "agentic." The distinction matters for architecture decisions. • Adversarial test sets first: before building any conversational system, generate ~100 adversarial synthetic conversations (easy, confusing, prompt-injection) with expected outcomes. Hemal lost two weeks skipping this step. • Start with the simplest architecture (one agent, many tools), measure failure modes, and only add orchestrators/sub-agents when failure data justifies the complexity. • Loop engineering: run agents through repeated cycles of hypothesis → experiment → analyze → fix → retest, journaling every experiment to a markdown file so context survives compaction. Review early cycles yourself, then let it run autonomously overnight. • Use cheap Chinese models (GLM, DeepSeek, Qwen) for internal experimentation; reserve American models (GPT-5.5, Gemini) for production-facing or HIPAA-regulated work.

📝 SUMMARY This week's call featured a rotating show-and-tell covering personal agent harnesses, enterprise AI adaptation strategies, and mobile deployment tactics. Members demoed custom tools for tracking AI model usage across projects, automating subscription management, and running agentic workflows from mobile devices. Discussions ranged from rebuilding Codex functionality on banned networks to hiring philosophies for the AI-augmented era, with deep dives into cross-platform mobile stacks and biometric verification services. 💡 KEY INSIGHTS Scott Rippey demonstrated that hybrid-scanning local project folders alongside provider APIs can create a "Model Radar" to track which AI models are active versus deprecated across multiple client apps. He also showed how machine-to-machine sync via Bonjour keeps session databases consistent across multiple Macs. Ty Wells revealed an automation setup that switches between Claude and Codex subscriptions based on remaining usage percentage, preventing work loss from hitting quotas. His "intent capsules" concept—self-contained context bundles—allows cold agent sessions to execute plans without prior conversation history, enabling remote work continuation from anywhere including mobile devices. Patrick Chouinard explained how to recreate Codex review mechanics on GitHub Copilot CLI when OpenAI tools are banned at work, satisfying security requirements while maintaining access to frontier models. He emphasized that fully autonomous AI development is a misnomer because human intent is always required, and that goal-based automation only works with hard, testable facts rather than opinions. He also highlighted that just-in-time training via composable skills beats static courses, which become obsolete before completion. Daniel Zivkovic shared that requirements elicitation works better as an iterative "optometry" process (showing prototypes and asking "this or that") rather than upfront specification. He noted that running multiple AI-generated implementations in parallel causes conflicts, while sequential overnight runs create a morning review queue without collisions.

📝 SUMMARY This coaching call brought together founders and developers navigating the tension between product development and market validation. Brandon Hancock shared a candid business update on EMS SOAP pivoting its ideal customer profile from slow-moving fire departments to faster-closing private ambulance companies under cash-flow pressure. Patrick Chouinard detailed his enterprise Claude Code rollout expanding from 250 to 2,500 users, including a self-improving support knowledge base and automated training generation. The group also demoed diverse projects including Agent Task for cross-agent orchestration, a multi-model security review system, fully automated ERP pipelines, an AI photo booth, and school pickup management software. Throughout, Brandon coached on sales segmentation frameworks, disciplined outreach habits, and productizing internal tools while technical discussions covered cross-model review strategies, local model use cases, and AI architecture decisions. 💡 KEY INSIGHTS Whale/Minnow/Tuna Segmentation: Brandon introduced a sales framework distinguishing whales (huge slow contracts taking a year or more), minnows (small deals handled via automation), and tunas (mid-size $20-40k accounts closing in months). Most startups should focus limited selling time on tunas for sustainable growth. Cash-Flow Driven ICP Pivots: While standard advice suggests staying with one avatar until saturated, burning startups may need to pivot ICPs specifically to accelerate sales cycles when survival is at stake. Sales Enablement as Champion Creation: Providing prospects with case studies, ROI calculators, and one-pagers they can hand up their chain of command turns customers into internal advocates who sell to both department heads and budget authorities. Cross-Model Review Discipline: Never use the same model family to review code it generated. Using Claude for generation and Codex (or another vendor) for review surfaces different cognitive biases and catches issues the original model misses.

With July 7th coming up, I’ve been thinking less about using Fable and more about capturing its value. My approach has been to spend my remaining time with Fable on work that only Fable can realistically do. Not writing code. Not debugging. Not implementing features. Instead, I’ve been asking it to design systems, challenge architectures, and produce implementation specifications for projects that are simply too large or too interconnected for me to have tackled with previous models. The key is this: The output isn’t the final product. The output is a specification detailed enough that Opus, GPT-5.5, Codex, or another implementation model can execute it later. In other words, I’m using Fable to build my implementation backlog. By the time July 7 arrives, I don’t want to have “used” Fable for a few conversations. I want to have months of high-quality architectural work waiting to be implemented. It’s almost like borrowing the brain of a senior architect for a limited time, then leaving the office with a stack of blueprints your engineering team can build from over the coming months. If you still have access, I’d encourage you to spend less time asking it to do the work and more time asking it to define the work. That knowledge doesn’t disappear when access ends. In many cases, it’s the most durable asset you’ll get from the entire preview.